Jamie Joyce: Scaled Democratic Reasoning | Intelligence at the Frontier 2026

[music]

Especially since today I'm going to present to you a really, really complex argument that I have been making for years but is becoming more salient over time. So, I'm really excited to get started with that. That concept that I'm going to be arguing for is something new that we've coined and we're going to be releasing a paper on it, but it's called scale democratic reasoning. And I'm just going to make one argument about it in very short period of time. It's also going to be a lot of information, so go ahead and let's get started.

Okay. Okay. So, I'm the founder of the Society Library. You haven't heard about our organization, our interest is in raising epistemic standards and in creating the capabilities to meet those standards. And this means standards in truth seeking and deliberation as well as creating like technologies to enable scale deliberation.

Go ahead. Okay, and the standard that is new that I'm talking about is scale democratic reasoning. Now, more specifically about the organization is really quick. We build many epistemic tools that are actually configurable, modularized. They combine different stakeholder needs and what is the output of the combination is a bunch of products and tools.

Go ahead. We're actually put all that in a dashboard that's going to be publicly available, so look out for that. And then the types of things that are available that we worked on in the past have been things like interactive intelligence briefs, debate prediction, AI agent search and index building, fact checking, even writing legislation and voting protocols. And these are things that we've done in the past, but ultimately all of this R&D has been working up to this idea of scale democratic reasoning. Now, are you ready for me to make the argument for democratic scale reasoning?

Yes? Okay, it's going to be a lot. It's going to be really long, so get ready for it, okay? The amount of knowledge relevant to a civilization scale debate and the complexity of the logical reasoning likely to occur in such a debate is beyond human reasoning capabilities both individually and in groups even with AI augmentation so that it cannot result in a fully informed logical decision appropriately caveated with an acknowledgement and awareness of uncertainty. Got it?

Okay, great. A little bit more. Essentially, cuz sometimes you got to repeat. I really want I really want this to catch. Essentially, the computation available to the human mind cannot grasp highly complex issues reliably enough.

And funnily enough, even if the maximally contextual available information about a civilization scale issue is readily available enough to make a fully informed decision if it is grounded in appreciation for inquiry into truth from an intellectually humble point of view in a Western logical modality which would therefore necessitate debate. Got it? Yeah, who agrees? Okay, let's go.

[laughter]

So, because I'm talking to a lot of people who are in the deliberative democracy space and deliberation tech space, immediately when I make that argument, I get all of these hesitations. So, let me just quickly clarify. What I am not saying is that the Western logical modalities are the best way of reasoning such as deductive reasoning. I am not saying that other epistemologies for knowing are invalid. I am not saying that other methods of deliberation are not useful to perform, that people cannot arrive at truth about civilization scale debate and understanding, that aligning with truth is even the goal of political reasoning, or that logical debate excludes moral reasoning, values, or spiritual knowledge.

It certainly doesn't. Literally anything that can be put into language, anything that is evitable, can be represented in a deliberation, especially the takes of rest which I'm about to show you. So, it doesn't exclude any tradition of ways of knowing as long as could be put into words, it can fit into scale down your quadratic reasoning. Even this right here, this uh divination practice. What is being derived from this divination can be put into words.

Those words can be represented in scaled down democratic reasoning. So, that's the That's the pitfall of people grasping this argument. Um and like literally even these database structures are what I'm about to show you could go into pictures, videos, cartoons, anything describable language. Go ahead. All right.

So, I'm going to make the argument one more time, but very simply, and I'm going to check in and maybe it will I'll get some true believers here. Basically, I'm going to use more simple language. Many political issues are very big and complex, and people don't necessarily have the time or attention to make sense of it all if you want to be maximally democratic and hear from all points of view. Therefore, we argue that a new form of computation for democratic reasoning beyond human-based voting mechanisms is both warranted and could also be viable. How does that sound to anyone?

Okay, we get some We get some All right, we got some nods and fingers up. All right, that's great. Uh three times is the charm. Thank you. Okay.

Now, one of the reasons why I make this argument is because there's been a lot of virtue around human working memory and the like how much information people can hold in their minds. Um according to uh one paper, it is approximately four items that can be held in working memory simultaneously and it sometimes can squeak up to seven if they are chunked effectively, but it is argued that these constraints cannot be overcome through education or intentional effort. And in general, it limits how humans can engage with the complexity of policy domains. The really important thing to know if um you are uh accepting of like a tradition of you know, evaluating evidence as being like potentially verifiable from neuroscience is that certainty is a feeling. So, when something is salient, something makes sense, we're like, "Yeah, that's That's not necessarily a deductive process in our brain that we get that feeling from having earned doing this logical deductive work.

It is literally just a feeling. It could be from divination, it could be from looking at the stars, it could be from reading, it could be from so many things. So, when we're talking about like having to hold so much in our minds when we as citizens are democratically reasoning, let's take climate policy as an example. The climate policy question not only engages in atmospheric science such as carbon budgets and tipping points, but energy systems such as grid engineering, storage technology, economics like carbon pricing, transition costing, equity like distributive justice, developmental rights, and geopolitics that include technology transfer, international cooperation, etc. So, each one of these domains has hundreds and hundreds and hundreds of claims that you would need to adjudicate.

So, to imagine you could just take any person who has maybe some fraction of knowledge about this issue and throw enough of them in the room that they then would have the ability to logically reason about this at scale is something I'm arguing as needing not the most optimal way. Instead, we could look towards technology to help scale what is called scaled deliberative democracy. Okay. All right. Now, if any of you have seen a talk that I did before, you've probably seen these graphs.

Has anyone seen this before? Okay. Oh, sweet. Very few. Great.

I'm glad I included this. So, this is a deliberation graph. Basically, this is a visual representation of like this nested hierarchical very dense structure of arguments, claims, and evidence that we derive from old Anglo artifacts. So, when mapping a debate, we look across every form of media you can imagine from documentaries and TikTok and social media to scholarly articles, economic impact assessments, whatever. Extract the arguments, ground it, maintain the chain of provenance, definitions.

Each one of these little nodes is very dense and nested with information. So, what does that actually mean to reason from the logical Western modality? Again, not the only way, but interestingly very a encompassing because it can encompass language itself. Go ahead. So, here is just a little snippet.

And this is just one topic, one question. This particular graph has nine different positions, but I'm only showing one. So, we have one position. This particular position has like I think 15 different categories, which let's just say flavors of reasoning. This one is has to do with safety and well-being.

And this is just one argument. So, this is a tiny little sliver of a little graph. And you can see that in order for this argument to be deductively true right here, every single one of these premises we need to be true. So, you have to argue about the truth of these premises using reasoning and evidence. Go ahead.

Now, what does that look like? When you are arguing about whether each one of these premises are true, this is the graph. Uh it's, you know, a lot of little tiny slivers if you can discern what that is. Probably hundreds and hundreds of arguments going down many layers arguing about whether these premises that this argument relies on in order to be true can be true. And I think I forgot to Go ahead.

Move forward. So, I think I forgot to add the actual video. But this is a video that's very impressive that scrolls by very fast showing that that uh there's many, many, many arguments across this entire graph when we unpacked just three layers. So, when I play it it's like 2 minutes long. It's like my voice.

Just imagine so many nodes going by, all of them be deduplicated, completely relevant uh artifact-based arguments about this one issue having to do with the nuclear power plant. And this nuclear power plant issue was a state-level issue. It's not talking about climate change. We're not talking about the big global issues. Just one state-level issue ended up being 5,862 points.

So, the human mind can only keep like four to seven-ish points. And this alone was a one one small debate at a lowly local nuclear power plant is 5,862 points going ranging across the economics to the seismicity, to the ethics, indigenous rights, all these other things. What I'm arguing is that we probably need technology to help us to scale reasoning so that we can actually derive policy outcomes in a way that's going to be maximally inclusive of all these different points of view and also has a process of helping us to adjudicate what may be in accordance with reality, which of course is up for debate across different epistemological modalities. Go for it. Yay, okay.

Now, when we talk about this in this space in spaces like these with a lot of deliberation tech, for some reason even as I make these arguments and I show this data, these graphs, there's still this kind of like hesitation of like no, there's something about democratic reasoning that is inherently human. We want to be in the room. We want to be in these citizen assemblies. I personally never do. Um I like to be at home behind the computer doing this.

But uh again, not saying that those forms of engaging with the democratic process are invalid cuz they're certainly not. Very useful elicitation, very useful for civic spirit and things like this and depolarization. But I want uh I would love for this space to also try on that there's been so many different discoveries and new apertures for perceiving the world that are only possible with technology. When we're talking about um like protein structure prediction, antibiotic discovery, AlphaGeometry 2 from DeepMind solving math, like if we think about something as simple as a microscope and a telescope compared to our vision, we can only see so far, we can only see so deep. It's okay to use technology to augment our capabilities.

And like in so far as I know, it's not on the timeline that we're going to install microscopes in our eyes and truly be augmented with that. So it's okay to use external technology in order to extend our capabilities. And I think one of the most important capabilities that we do need to extend and make auditable, super inclusive, very grounded in evidence if it exists, is something like the decisions that we make to govern the resources uh of our civilization and make choices that are going to affect the future of all humans alive today. So that is my argument. Go ahead and go to next line.

Um we're going to try writing this all out in two columns and writing a paper to really try to make that argument. So, um stay tuned for that. I'm also incredibly open to hearing any counter arguments you all have and I just really thank you viewers. I'm thanks to Mario for having me.

[music]